Schwinger-Keldysh-Galley Action Principle

• The Schwinger-Keldysh-Galley action principle is a framework that doubles degrees of freedom to convert traditional two-endpoint boundary problems into initial-value formulations.
• It originates from the closed-time-path formalism, using forward/backward and average/difference variables to capture both conservative dynamics and nonconservative effects like dissipation.
• The formulation underlies effective actions in fields such as hydrodynamics and holography, linking fluctuation-dissipation relations with real-time transport phenomena.

The Schwinger–Keldysh–Galley double-variable action principle is a variational framework for real-time dynamics in which every degree of freedom is doubled, either as forward/backward contour variables $(q_1,q_2)$ or as average/difference variables $(q_+,q_-)$. In the conservative point-particle setting, Horowitz and Rothkopf derive this structure directly from the quantum Schwinger–Keldysh in–in path integral and show that its classical limit yields an action principle for initial-value problems rather than the standard two-endpoint boundary-value problem of Hamilton’s principle. Related formulations extend the doubled action to nonconservative Hamiltonian mechanics and to dissipative effective actions in hydrodynamics and holography (Horowitz et al., 2 Mar 2026, Aykroyd et al., 23 Jul 2025, Bu et al., 2020).

1. Closed-time-path origin and doubled variables

In nonrelativistic quantum mechanics with Hamiltonian

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$

the in–in expectation value of the coordinate operator at a final time $t_f$, for an initial pure Gaussian wave packet $|\psi(t_i)\rangle$ peaked at $(x_0,p_0)$, is written as

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$

Trotter expansion of the forward and backward evolution operators, together with insertions of complete sets of $|x\rangle$ and $|p\rangle$ eigenstates on a closed time contour, produces a discretized path integral with two branches, conventionally labelled $1$ and $(q_+,q_-)$0, and phase

$(q_+,q_-)$1

where

$(q_+,q_-)$2

The classical limit is then extracted by stationary phase while retaining the full phase-space measure, including the initial Gaussian smearing and the final $(q_+,q_-)$3-constraint (Horowitz et al., 2 Mar 2026).

The same closed-time-path structure appears in field theory. One introduces two copies of each dynamical field, $(q_+,q_-)$4 and $(q_+,q_-)$5, or equivalently the Keldysh variables

$(q_+,q_-)$6

In this notation the physical, or classical, solutions satisfy $(q_+,q_-)$7, while $(q_+,q_-)$8-variations generate response functions and noise. In the diffusion effective-action context, the doubling is the mechanism that captures dissipation and fluctuations in real time and enforces fluctuation–dissipation and unitarity (Bu et al., 2020).

2. Doubled action in average/difference form

For the point-particle derivation, it is convenient to pass from the two contour branches to average and difference variables at each time slice,

$(q_+,q_-)$9

and similarly

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$0

After trading the final $\hat H=\frac{\hat p^2}{2m}+V(\hat x),$1 and the initial Gaussian factor for Lagrange multipliers, the discrete Schwinger–Keldysh–Galley action becomes

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$2

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$3

In the continuum limit,

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$4

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$5

Integrating out $\hat H=\frac{\hat p^2}{2m}+V(\hat x),$6 yields the configuration-space form

$\hat H=\frac{\hat p^2}{2m}+V(\hat x),$7

with $\hat H=\frac{\hat p^2}{2m}+V(\hat x),$8 and $\hat H=\frac{\hat p^2}{2m}+V(\hat x),$9 (Horowitz et al., 2 Mar 2026).

For nonconservative systems, the doubled action is enlarged by an antisymmetric coupling $t_f$0,

$t_f$1

This separates the doubled Lagrangian into conservative terms $t_f$2 and a nonconservative sector carried entirely by $t_f$3 (Aykroyd et al., 23 Jul 2025).

3. Variational structure and initial-value conditions

The defining feature of the doubled variational principle is its boundary prescription. In the conservative point-particle formulation one varies $t_f$4 and $t_f$5 independently while imposing only

$t_f$6

and leaving $t_f$7 and $t_f$8 free. Stationarity then yields, at the initial time,

$t_f$9

in the bulk,

$|\psi(t_i)\rangle$0

and at the final time,

$|\psi(t_i)\rangle$1

This is the precise mechanism by which the doubled action converts the standard endpoint-fixed variational problem into an initial-value problem for the $|\psi(t_i)\rangle$2 trajectory (Horowitz et al., 2 Mar 2026).

In the nonconservative formulation the corresponding initial-value prescription is stated in terms of the two copies. At $|\psi(t_i)\rangle$3 one fixes

$|\psi(t_i)\rangle$4

while at $|\psi(t_i)\rangle$5 one imposes only the free copy-matching conditions

$|\psi(t_i)\rangle$6

Independent variation gives two coupled Euler–Lagrange equations, and because the two equations are related by exchanging the labels $|\psi(t_i)\rangle$7 and the initial-value problem is unique, the on-shell solution obeys

$|\psi(t_i)\rangle$8

throughout the interval. This is the physical-limit trajectory of the doubled system (Aykroyd et al., 23 Jul 2025).

4. Minus-path dynamics, backward propagation, and normalization

A recurrent misconception is that the minus path merely parametrizes fluctuations around a single classical solution carried by the plus path. In the analysis of Horowitz and Rothkopf, both the plus and minus Keldysh paths have classical paths and fluctuations in the full path integral, and the fluctuations of both are required for the correct normalization of the classical limit (Horowitz et al., 2 Mar 2026).

The Euler–Lagrange equation for $|\psi(t_i)\rangle$9 becomes homogeneous when $(x_0,p_0)$0, and it is supplemented by the single boundary condition $(x_0,p_0)$1. Picard–Lindelöf then implies the unique solution

$(x_0,p_0)$2

In the discrete description, the same conclusion follows from backward-in-time difference equations for $(x_0,p_0)$3 and any associated $(x_0,p_0)$4, starting from $(x_0,p_0)$5 and $(x_0,p_0)$6, whose unique solution is $(x_0,p_0)$7. The minus-path solution therefore propagates backward in time, and one does not need to impose $(x_0,p_0)$8 by hand when taking the classical limit of the Schwinger–Keldysh formalism (Horowitz et al., 2 Mar 2026).

The role of the minus path is nonetheless essential before the classical saddle is taken. It enforces the final $(x_0,p_0)$9-constraint joining the two Schwinger–Keldysh branches, contributes the Gaussian fluctuations required for normalization, and only after variation and stationary phase is it dynamically driven to the trivial solution. This is the central resolution of the initial-value puzzle in the doubled formalism (Horowitz et al., 2 Mar 2026).

5. Nonconservative mechanics, Hamiltonian extension, and gauge freedom

The doubled-variable action admits a Hamiltonian formulation for nonconservative systems. Writing

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$0

and introducing

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$1

antisymmetry under $$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$2 implies that $$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$3 contains only odd powers of $$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$4. Linearization around the physical limit gives

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$5

with $$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$6 and a nonconservative generalized force

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$7

The physical-limit equation of motion becomes

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$8

This decomposition isolates the conservative and dissipative sectors of the doubled Lagrangian (Aykroyd et al., 23 Jul 2025).

The doubled canonical momenta are

$$\langle \hat x\rangle(t_f)=\langle\psi(t_i)|\,U^\dagger(t_f,t_i)\,\hat x\,U(t_f,t_i)\,|\psi(t_i)\rangle.$$9

and the Legendre transform defines the extended Hamiltonian

$|x\rangle$0

A key structural result is a gauge freedom in the physical momentum:

$|x\rangle$1

accompanied by

$|x\rangle$2

with

$|x\rangle$3

The physical equation of motion is unchanged, and the Hamiltonian language produces a family of gauge-related nonconservative Hamiltonians that are canonically equivalent on the physical slice $|x\rangle$4, $|x\rangle$5. A convenient conservative gauge is

$|x\rangle$6

which restores the standard definition $|x\rangle$7 (Aykroyd et al., 23 Jul 2025).

The same work shows that any second-order ODE system

$|x\rangle$8

can be embedded in the enlarged symplectic manifold by choosing an invertible momentum map $|x\rangle$9, defining $|p\rangle$0, and constructing the linearized Lagrangian

$|p\rangle$1

together with the Lie-form Hamiltonian

$|p\rangle$2

In this linear form,

$|p\rangle$3

and the extended Poisson bracket with $|p\rangle$4 reproduces the original nonconservative dynamics when restricted to $|p\rangle$5 (Aykroyd et al., 23 Jul 2025).

6. Effective actions, hydrodynamics, and holographic realization

In dissipative field theory the Schwinger–Keldysh–Galley doubling organizes the effective action for low-frequency modes. For charge diffusion, the generic quadratic effective action is written as

$|p\rangle$6

where $|p\rangle$7 is the physical background gauge potential and $|p\rangle$8 its Schwinger–Keldysh conjugate source. The operators

$|p\rangle$9

encode dissipation and noise, respectively, while KMS symmetry enforces

$1$0

In this setting the doubled variables are not auxiliary bookkeeping devices; they generate the response sector and the stochastic sector simultaneously (Bu et al., 2020).

A holographic realization uses two copies of a Schwarzschild–AdS$1$1 black brane and two Maxwell fields, $1$2 and $1$3, filling in the thermal Schwinger–Keldysh contour. The bulk action is

$1$4

with the relative minus sign implementing the contour orientation. Dirichlet conditions are imposed on the two asymptotic boundaries, and regularity at the horizon together with matching around a small Euclidean cap uniquely fixes the bulk solution in terms of the boundary data. The on-shell action then reduces to the doubled effective action above, exact to all orders in the derivative expansion (Bu et al., 2020).

Variation with respect to the Schwinger–Keldysh sources yields two currents,

$1$5

and the classical hydrodynamic current satisfies an all-order constitutive relation. Allowing $1$6 introduces a stochastic term, and rewriting the response field in terms of a physical noise variable $1$7 gives a Gaussian noise action with correlator

$1$8

Because $1$9 depends nontrivially on $(q_+,q_-)$00 and $(q_+,q_-)$01, the noise is coloured and generically non-local in time and space. This field-theoretic realization shows that the same doubled action principle that yields classical initial-value mechanics also underlies dissipative transport, fluctuation–dissipation relations, and real-time holography (Bu et al., 2020).